210 likes | 326 Views
Using Selection Criteria to Optimize Analysis in High Energy Physics. Comparing Methods to Find New Particles. Chris Davis*, Dr. David Toback , Daniel Cruz Texas A&M University Dr. Joel Walker, Jacob Hill Sam Houston State University. Outline. Overview
E N D
Using Selection Criteria to Optimize Analysis in High Energy Physics • Comparing Methods to Find New Particles Chris Davis*, Dr. David Toback, Daniel Cruz Texas A&M University Dr. Joel Walker, Jacob Hill Sam Houston State University
Outline • Overview • Motivation for using selection criteria to find new particles • Using Selection Criteria (Cuts) • Comparing Different Approaches • Results
Motivation • We want to be more sensitive to new particles in High Energy Physics • Huge amount of collisions at colliders such as LHC means lots of data to look through • Many ways to look for new particles • However, most are dominated by Standard Model particle “backgrounds” • In some places, new particle “signal” dominates the background • Using selection criteria allows us to be the most sensitive to new particles in these regions
The Data We Used Hypothesis 1 • Data is from a diphoton search for supersymmetry at Fermilab1 • A typical, simple search involves counting • Number of Background events expected • Number of Signal events expected • How many events are observed in the experiment • Add up observed events to determine which hypothesis is more consistent with data Hypothesis 2 1. Eunsin Lee, TAMU Ph.D. Thesis (2010), PRL 104
Selection Criteria • Selection criteria are used to optimize searches • Select only events that pass certain criteria • New particles easily pass them • Few Background events also pass Thrown out
Single Selection Criterion • Creates a single set of data starting at A • Throw out all events that do not pass our criterion, count events from A→∞ • Lowering the value of A adds in more background, more signal • Raising value of A takes out background, but also signal • We look at data that is most sensitive to signal A Thrown out
Single Criterion in Experiment • Cross section, σ, is a measure of sensitivity • Lower σ, better sensitivity • Higher σ, worse sensitivity • Vary A to optimize sensitivity • Can we get better sensitivity by doing a more sophisticated analysis? Not much signal here so poor sensitivity Lots of background here so poor sensitivity Best balance between signal and background, Best sensitivity 8
Two Selection Criteria • Data is placed into two sets • Count events from A→B and B→∞ • This is a more sophisticated analysis • Does being more sophisticated translate to being more sensitive? • Systematic errors can be introduced, we’ll deal with the simplest case without them in this talk A B Thrown out
Main Question • Is it better to do one or two separate sets of independent criteria? • If we use two selection criteria, can we become more sensitive to new particles?.....Yes, will show! • Is using two selection criteria always more sensitive than using a single selection criterion?.....Surprisingly no, will show! OR
Two Selection Criteria in Experiment • Optimal criteria give lower σ than the optimal single criterion • ≈5% less in this particular experiment • More sensitive! • Varying A and B to optimize sensitivity Minimum A≥B
Can it be Worse? B Fixed A Varied • Look at two criteria in one dimension to compare with an optimal single criterion • Arbitrarily fix B and vary A • There is a region where two criteria are better • However, also regions where two criteria are worse! Worse! Better!
Conclusions • Our sensitivity to new particles is improved when we use selection criteria • We have determined that • Two criteria CAN be better than a single, optimized criterion • Need to look for a minimum! • Two criteria CAN ALSOgive a worse result if used incorrectly
Limit Calculator • Example One-Cut input • 160 • 1 • -1 2.59 .0790 .1218 4.251 .3188 • 0 • Example Two-Cut Input • 360 • 2 • -1 2.59 .0399 .1218 4.218 .3188 • -1 2.59 .0391 .1218 .0326 .3188 • 0
Expected Cross Sections • Nevents = Luminosity * σproduction * Acceptance • Find 95% confidence limits on σproduction • Taking cuts allows us to optimize expected σ • Used improved Limit Calculating program1 1. Developed by Dr. Joel Walker, Sam Houston State University
Splitting Single Cut into Two • If you take a single cut and place a B cut in it, you will always improve your sensitivity • Possibly not much better, but never worse
Binned Value Two-Cut • The optimal cuts give 20.1 fb at A:240 B:360
Percentage Decrease • Two cuts can be used to improve the optimal expected limit • Able to achieve slightly under 10% decrease (8.64%)
Acceptance • Related to signal by a scaling factor